Research On Traffic Target Sensing Algorithm Based On Radar And Vision Fusion

In the field of intelligent transportation perception,radar and camera are two commonly used sensors in order to obtain information such as the location,category,and speed of traffic targets.Due to the different perception capabilities of radar and camera in different aspects,the solution based on radar and vision fusion has gradually become a mainstream solution.However,most of the current radar and vision fusion solutions are not designed with the core goal of improving target estimation accuracy.The research objective of this article is to design a new radar and vision fusion architecture,study its key issues,and achieve more accurate detection and perception of traffic targets.In the process of radar and vision fusion,three key issues are mainly involved: radar and vision spatiotemporal synchronization,radar and vision measurement association,and radar and vision estimation fusion.Thesis will conduct research on each of them.(1)Aiming at the problem of radar and vision spatiotemporal synchronization,inspired by the visual monocular ranging algorithm based on road geometric models,thesis proposes a radar and vision spatial synchronization scheme based on visual monocular ranging,which is more realistic and convenient for subsequent processing steps than the model based spatial synchronization method;In addition,thesis improves the interpolation and extrapolation method to achieve time synchronization between radar and vision.(2)Aiming at the association problem between radar and vision measurements,thesis proposes an association algorithm based on visual classification results,which carefully designs the size,position,and shape of the association frame.This algorithm has higher association accuracy than the fixed threshold association algorithm;In addition,thesis draws on evaluation indicators in the field of computer vision and proposes evaluation indicators to measure the quality of the association between radar and visual measurements.(3)Aiming at the estimation fusion of radar and vision,thesis proposes a modeling theory for visual measurement distance(VMD)uncertainty,introducing visual measurement into statistical models;Moreover,based on the VMD theoretical model,thesis improves the fusion algorithm based on covariance crossover through dynamic parameter design,which is more suitable for radar and vision fusion scenarios.All the above work has been verified by actual data,and ultimately achieved more accurate perception of traffic targets based on radar and vision fusion.It can be applied to practical scenarios such as road monitoring and intelligent transportation in the future.